Instructions for locally setting up and running Wan 2.2 on a Windows-based computer: Detailed walkthrough
Wan 2.2, an open-source AI model developed by Alibaba's Institute for Intelligent Computing, is making waves in the world of artificial intelligence. This powerful tool, designed to generate videos directly from text or images, can now be run locally on your Windows PC, offering benefits such as no usage limits, faster feedback, privacy, and customization.
To get started, you'll need to meet some minimum system requirements. These include Windows 10 or 11 (64-bit), an NVIDIA GPU with CUDA support, 8 GB VRAM (RTX 3060 or higher recommended), 16 GB RAM minimum, 20-100 GB free storage, and the latest NVIDIA GPU drivers and CUDA toolkit installed.
Wan 2.2 runs inside ComfyUI, a modular graphical interface built for AI workflows. You can download ComfyUI from its GitHub page or website, extract, and install it on your system. Once installed, you can input a text prompt, adjust settings like resolution and frame count, and hit Run to generate an AI-created video locally.
For beginners, SwarmUI might be a gentler entry point into the Wan 2.2 ecosystem. SwarmUI is a community-built wrapper for ComfyUI that simplifies the installation process. It automatically downloads models, configures workflows, and lets users prompt and render without worrying about file paths or dependencies.
Wan 2.2 offers various options within ComfyUI, including T2V-A14B, I2V-A14B, and TI2V-5B. If your graphics card is less than a top-tier RTX, it's recommended to start with TI2V-5B, a lightweight text-to-image-to-video hybrid workflow designed for mid-range GPUs.
Running Wan 2.2 locally offers a unique experience. The first clip rendered using Wan 2.2 might be short and imperfect, but it's yours - born from your words, generated by your machine, and limited only by your imagination. Examples of Wan 2.2's creations include neon-soaked cityscapes, surreal dream montages, and futuristic characters.
However, there might be instances where you encounter issues such as Out of Memory (OOM) errors or extremely slow rendering. In such cases, reducing resolution, frame count, or switching to TI2V-5B can help. Additionally, using SSD storage, closing background apps, or lowering parameters can improve rendering speed.
Updating GPU drivers and ensuring CUDA toolkit matches the PyTorch build can help resolve CUDA errors or driver mismatches. The model weights, the massive trained files that power Wan 2.2, need to be downloaded and placed in the models folder inside ComfyUI.
Wan 2.2 is released as open-source under the Apache 2.0 license. The latest tutorials show how to run Wan 2.2 locally on a Windows PC, making it accessible for anyone interested in exploring the exciting world of AI-generated videos.
